claudecode-subagent-system
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a subagent coordination system that processes natural language tasks, establishing a potential attack surface for indirect prompt injection. Ingestion points: Task strings accepted by the
codex subagent delegatecommand and theroute_taskPython method in SKILL.md. Boundary markers: The configuration examples and illustrative code lack explicit delimiters or instructions to isolate potentially malicious embedded commands within task inputs. Capability inventory: The system schema supports subagents with significant tool access (e.g., 'security-scanner') and file permissions (e.g., 'read-code', 'write-reports') in SKILL.md. Sanitization: The provided implementation examples do not include input validation or filtering logic for the task data being routed.
Audit Metadata